locks: allow lockd to process blocked locks during grace period
[safe/jmp/linux-2.6] / fs / lockd / svc.c
1 /*
2  * linux/fs/lockd/svc.c
3  *
4  * This is the central lockd service.
5  *
6  * FIXME: Separate the lockd NFS server functionality from the lockd NFS
7  *        client functionality. Oh why didn't Sun create two separate
8  *        services in the first place?
9  *
10  * Authors:     Olaf Kirch (okir@monad.swb.de)
11  *
12  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
13  */
14
15 #include <linux/module.h>
16 #include <linux/init.h>
17 #include <linux/sysctl.h>
18 #include <linux/moduleparam.h>
19
20 #include <linux/sched.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/uio.h>
24 #include <linux/slab.h>
25 #include <linux/smp.h>
26 #include <linux/smp_lock.h>
27 #include <linux/mutex.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30
31 #include <linux/sunrpc/types.h>
32 #include <linux/sunrpc/stats.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/svc.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <net/ip.h>
37 #include <linux/lockd/lockd.h>
38 #include <linux/lockd/sm_inter.h>
39 #include <linux/nfs.h>
40
41 #define NLMDBG_FACILITY         NLMDBG_SVC
42 #define LOCKD_BUFSIZE           (1024 + NLMSVC_XDRSIZE)
43 #define ALLOWED_SIGS            (sigmask(SIGKILL))
44
45 static struct svc_program       nlmsvc_program;
46
47 struct nlmsvc_binding *         nlmsvc_ops;
48 EXPORT_SYMBOL(nlmsvc_ops);
49
50 static DEFINE_MUTEX(nlmsvc_mutex);
51 static unsigned int             nlmsvc_users;
52 static struct task_struct       *nlmsvc_task;
53 static struct svc_rqst          *nlmsvc_rqst;
54 int                             nlmsvc_grace_period;
55 unsigned long                   nlmsvc_timeout;
56
57 /*
58  * These can be set at insmod time (useful for NFS as root filesystem),
59  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
60  */
61 static unsigned long            nlm_grace_period;
62 static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
63 static int                      nlm_udpport, nlm_tcpport;
64 int                             nsm_use_hostnames = 0;
65
66 /*
67  * Constants needed for the sysctl interface.
68  */
69 static const unsigned long      nlm_grace_period_min = 0;
70 static const unsigned long      nlm_grace_period_max = 240;
71 static const unsigned long      nlm_timeout_min = 3;
72 static const unsigned long      nlm_timeout_max = 20;
73 static const int                nlm_port_min = 0, nlm_port_max = 65535;
74
75 #ifdef CONFIG_SYSCTL
76 static struct ctl_table_header * nlm_sysctl_table;
77 #endif
78
79 static unsigned long get_lockd_grace_period(void)
80 {
81         /* Note: nlm_timeout should always be nonzero */
82         if (nlm_grace_period)
83                 return roundup(nlm_grace_period, nlm_timeout) * HZ;
84         else
85                 return nlm_timeout * 5 * HZ;
86 }
87
88 unsigned long get_nfs_grace_period(void)
89 {
90         unsigned long lockdgrace = get_lockd_grace_period();
91         unsigned long nfsdgrace = 0;
92
93         if (nlmsvc_ops)
94                 nfsdgrace = nlmsvc_ops->get_grace_period();
95
96         return max(lockdgrace, nfsdgrace);
97 }
98 EXPORT_SYMBOL(get_nfs_grace_period);
99
100 static unsigned long set_grace_period(void)
101 {
102         nlmsvc_grace_period = 1;
103         return get_nfs_grace_period() + jiffies;
104 }
105
106 static inline void clear_grace_period(void)
107 {
108         nlmsvc_grace_period = 0;
109 }
110
111 /*
112  * This is the lockd kernel thread
113  */
114 static int
115 lockd(void *vrqstp)
116 {
117         int             err = 0, preverr = 0;
118         struct svc_rqst *rqstp = vrqstp;
119         unsigned long grace_period_expire;
120
121         /* try_to_freeze() is called from svc_recv() */
122         set_freezable();
123
124         /* Allow SIGKILL to tell lockd to drop all of its locks */
125         allow_signal(SIGKILL);
126
127         dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
128
129         /*
130          * FIXME: it would be nice if lockd didn't spend its entire life
131          * running under the BKL. At the very least, it would be good to
132          * have someone clarify what it's intended to protect here. I've
133          * seen some handwavy posts about posix locking needing to be
134          * done under the BKL, but it's far from clear.
135          */
136         lock_kernel();
137
138         if (!nlm_timeout)
139                 nlm_timeout = LOCKD_DFLT_TIMEO;
140         nlmsvc_timeout = nlm_timeout * HZ;
141
142         grace_period_expire = set_grace_period();
143
144         /*
145          * The main request loop. We don't terminate until the last
146          * NFS mount or NFS daemon has gone away.
147          */
148         while (!kthread_should_stop()) {
149                 long timeout = MAX_SCHEDULE_TIMEOUT;
150                 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
151
152                 if (signalled()) {
153                         flush_signals(current);
154                         if (nlmsvc_ops) {
155                                 nlmsvc_invalidate_all();
156                                 grace_period_expire = set_grace_period();
157                         }
158                         continue;
159                 }
160
161                 timeout = nlmsvc_retry_blocked();
162
163                 if (time_before(grace_period_expire, jiffies))
164                         clear_grace_period();
165
166                 /*
167                  * Find a socket with data available and call its
168                  * recvfrom routine.
169                  */
170                 err = svc_recv(rqstp, timeout);
171                 if (err == -EAGAIN || err == -EINTR) {
172                         preverr = err;
173                         continue;
174                 }
175                 if (err < 0) {
176                         if (err != preverr) {
177                                 printk(KERN_WARNING "%s: unexpected error "
178                                         "from svc_recv (%d)\n", __func__, err);
179                                 preverr = err;
180                         }
181                         schedule_timeout_interruptible(HZ);
182                         continue;
183                 }
184                 preverr = err;
185
186                 dprintk("lockd: request from %s\n",
187                                 svc_print_addr(rqstp, buf, sizeof(buf)));
188
189                 svc_process(rqstp);
190         }
191         flush_signals(current);
192         if (nlmsvc_ops)
193                 nlmsvc_invalidate_all();
194         nlm_shutdown_hosts();
195         unlock_kernel();
196         return 0;
197 }
198
199 /*
200  * Make any sockets that are needed but not present.
201  * If nlm_udpport or nlm_tcpport were set as module
202  * options, make those sockets unconditionally
203  */
204 static int make_socks(struct svc_serv *serv, int proto)
205 {
206         static int warned;
207         struct svc_xprt *xprt;
208         int err = 0;
209
210         if (proto == IPPROTO_UDP || nlm_udpport) {
211                 xprt = svc_find_xprt(serv, "udp", 0, 0);
212                 if (!xprt)
213                         err = svc_create_xprt(serv, "udp", nlm_udpport,
214                                               SVC_SOCK_DEFAULTS);
215                 else
216                         svc_xprt_put(xprt);
217         }
218         if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport)) {
219                 xprt = svc_find_xprt(serv, "tcp", 0, 0);
220                 if (!xprt)
221                         err = svc_create_xprt(serv, "tcp", nlm_tcpport,
222                                               SVC_SOCK_DEFAULTS);
223                 else
224                         svc_xprt_put(xprt);
225         }
226         if (err >= 0) {
227                 warned = 0;
228                 err = 0;
229         } else if (warned++ == 0)
230                 printk(KERN_WARNING
231                        "lockd_up: makesock failed, error=%d\n", err);
232         return err;
233 }
234
235 /*
236  * Bring up the lockd process if it's not already up.
237  */
238 int
239 lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
240 {
241         struct svc_serv *serv;
242         int             error = 0;
243
244         mutex_lock(&nlmsvc_mutex);
245         /*
246          * Check whether we're already up and running.
247          */
248         if (nlmsvc_rqst) {
249                 if (proto)
250                         error = make_socks(nlmsvc_rqst->rq_server, proto);
251                 goto out;
252         }
253
254         /*
255          * Sanity check: if there's no pid,
256          * we should be the first user ...
257          */
258         if (nlmsvc_users)
259                 printk(KERN_WARNING
260                         "lockd_up: no pid, %d users??\n", nlmsvc_users);
261
262         error = -ENOMEM;
263         serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, AF_INET, NULL);
264         if (!serv) {
265                 printk(KERN_WARNING "lockd_up: create service failed\n");
266                 goto out;
267         }
268
269         if ((error = make_socks(serv, proto)) < 0)
270                 goto destroy_and_out;
271
272         /*
273          * Create the kernel thread and wait for it to start.
274          */
275         nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
276         if (IS_ERR(nlmsvc_rqst)) {
277                 error = PTR_ERR(nlmsvc_rqst);
278                 nlmsvc_rqst = NULL;
279                 printk(KERN_WARNING
280                         "lockd_up: svc_rqst allocation failed, error=%d\n",
281                         error);
282                 goto destroy_and_out;
283         }
284
285         svc_sock_update_bufs(serv);
286
287         nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
288         if (IS_ERR(nlmsvc_task)) {
289                 error = PTR_ERR(nlmsvc_task);
290                 svc_exit_thread(nlmsvc_rqst);
291                 nlmsvc_task = NULL;
292                 nlmsvc_rqst = NULL;
293                 printk(KERN_WARNING
294                         "lockd_up: kthread_run failed, error=%d\n", error);
295                 goto destroy_and_out;
296         }
297
298         /*
299          * Note: svc_serv structures have an initial use count of 1,
300          * so we exit through here on both success and failure.
301          */
302 destroy_and_out:
303         svc_destroy(serv);
304 out:
305         if (!error)
306                 nlmsvc_users++;
307         mutex_unlock(&nlmsvc_mutex);
308         return error;
309 }
310 EXPORT_SYMBOL(lockd_up);
311
312 /*
313  * Decrement the user count and bring down lockd if we're the last.
314  */
315 void
316 lockd_down(void)
317 {
318         mutex_lock(&nlmsvc_mutex);
319         if (nlmsvc_users) {
320                 if (--nlmsvc_users)
321                         goto out;
322         } else {
323                 printk(KERN_ERR "lockd_down: no users! task=%p\n",
324                         nlmsvc_task);
325                 BUG();
326         }
327
328         if (!nlmsvc_task) {
329                 printk(KERN_ERR "lockd_down: no lockd running.\n");
330                 BUG();
331         }
332         kthread_stop(nlmsvc_task);
333         svc_exit_thread(nlmsvc_rqst);
334         nlmsvc_task = NULL;
335         nlmsvc_rqst = NULL;
336 out:
337         mutex_unlock(&nlmsvc_mutex);
338 }
339 EXPORT_SYMBOL(lockd_down);
340
341 #ifdef CONFIG_SYSCTL
342
343 /*
344  * Sysctl parameters (same as module parameters, different interface).
345  */
346
347 static ctl_table nlm_sysctls[] = {
348         {
349                 .ctl_name       = CTL_UNNUMBERED,
350                 .procname       = "nlm_grace_period",
351                 .data           = &nlm_grace_period,
352                 .maxlen         = sizeof(unsigned long),
353                 .mode           = 0644,
354                 .proc_handler   = &proc_doulongvec_minmax,
355                 .extra1         = (unsigned long *) &nlm_grace_period_min,
356                 .extra2         = (unsigned long *) &nlm_grace_period_max,
357         },
358         {
359                 .ctl_name       = CTL_UNNUMBERED,
360                 .procname       = "nlm_timeout",
361                 .data           = &nlm_timeout,
362                 .maxlen         = sizeof(unsigned long),
363                 .mode           = 0644,
364                 .proc_handler   = &proc_doulongvec_minmax,
365                 .extra1         = (unsigned long *) &nlm_timeout_min,
366                 .extra2         = (unsigned long *) &nlm_timeout_max,
367         },
368         {
369                 .ctl_name       = CTL_UNNUMBERED,
370                 .procname       = "nlm_udpport",
371                 .data           = &nlm_udpport,
372                 .maxlen         = sizeof(int),
373                 .mode           = 0644,
374                 .proc_handler   = &proc_dointvec_minmax,
375                 .extra1         = (int *) &nlm_port_min,
376                 .extra2         = (int *) &nlm_port_max,
377         },
378         {
379                 .ctl_name       = CTL_UNNUMBERED,
380                 .procname       = "nlm_tcpport",
381                 .data           = &nlm_tcpport,
382                 .maxlen         = sizeof(int),
383                 .mode           = 0644,
384                 .proc_handler   = &proc_dointvec_minmax,
385                 .extra1         = (int *) &nlm_port_min,
386                 .extra2         = (int *) &nlm_port_max,
387         },
388         {
389                 .ctl_name       = CTL_UNNUMBERED,
390                 .procname       = "nsm_use_hostnames",
391                 .data           = &nsm_use_hostnames,
392                 .maxlen         = sizeof(int),
393                 .mode           = 0644,
394                 .proc_handler   = &proc_dointvec,
395         },
396         {
397                 .ctl_name       = CTL_UNNUMBERED,
398                 .procname       = "nsm_local_state",
399                 .data           = &nsm_local_state,
400                 .maxlen         = sizeof(int),
401                 .mode           = 0644,
402                 .proc_handler   = &proc_dointvec,
403         },
404         { .ctl_name = 0 }
405 };
406
407 static ctl_table nlm_sysctl_dir[] = {
408         {
409                 .ctl_name       = CTL_UNNUMBERED,
410                 .procname       = "nfs",
411                 .mode           = 0555,
412                 .child          = nlm_sysctls,
413         },
414         { .ctl_name = 0 }
415 };
416
417 static ctl_table nlm_sysctl_root[] = {
418         {
419                 .ctl_name       = CTL_FS,
420                 .procname       = "fs",
421                 .mode           = 0555,
422                 .child          = nlm_sysctl_dir,
423         },
424         { .ctl_name = 0 }
425 };
426
427 #endif  /* CONFIG_SYSCTL */
428
429 /*
430  * Module (and sysfs) parameters.
431  */
432
433 #define param_set_min_max(name, type, which_strtol, min, max)           \
434 static int param_set_##name(const char *val, struct kernel_param *kp)   \
435 {                                                                       \
436         char *endp;                                                     \
437         __typeof__(type) num = which_strtol(val, &endp, 0);             \
438         if (endp == val || *endp || num < (min) || num > (max))         \
439                 return -EINVAL;                                         \
440         *((int *) kp->arg) = num;                                       \
441         return 0;                                                       \
442 }
443
444 static inline int is_callback(u32 proc)
445 {
446         return proc == NLMPROC_GRANTED
447                 || proc == NLMPROC_GRANTED_MSG
448                 || proc == NLMPROC_TEST_RES
449                 || proc == NLMPROC_LOCK_RES
450                 || proc == NLMPROC_CANCEL_RES
451                 || proc == NLMPROC_UNLOCK_RES
452                 || proc == NLMPROC_NSM_NOTIFY;
453 }
454
455
456 static int lockd_authenticate(struct svc_rqst *rqstp)
457 {
458         rqstp->rq_client = NULL;
459         switch (rqstp->rq_authop->flavour) {
460                 case RPC_AUTH_NULL:
461                 case RPC_AUTH_UNIX:
462                         if (rqstp->rq_proc == 0)
463                                 return SVC_OK;
464                         if (is_callback(rqstp->rq_proc)) {
465                                 /* Leave it to individual procedures to
466                                  * call nlmsvc_lookup_host(rqstp)
467                                  */
468                                 return SVC_OK;
469                         }
470                         return svc_set_client(rqstp);
471         }
472         return SVC_DENIED;
473 }
474
475
476 param_set_min_max(port, int, simple_strtol, 0, 65535)
477 param_set_min_max(grace_period, unsigned long, simple_strtoul,
478                   nlm_grace_period_min, nlm_grace_period_max)
479 param_set_min_max(timeout, unsigned long, simple_strtoul,
480                   nlm_timeout_min, nlm_timeout_max)
481
482 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
483 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
484 MODULE_LICENSE("GPL");
485
486 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
487                   &nlm_grace_period, 0644);
488 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
489                   &nlm_timeout, 0644);
490 module_param_call(nlm_udpport, param_set_port, param_get_int,
491                   &nlm_udpport, 0644);
492 module_param_call(nlm_tcpport, param_set_port, param_get_int,
493                   &nlm_tcpport, 0644);
494 module_param(nsm_use_hostnames, bool, 0644);
495
496 /*
497  * Initialising and terminating the module.
498  */
499
500 static int __init init_nlm(void)
501 {
502 #ifdef CONFIG_SYSCTL
503         nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
504         return nlm_sysctl_table ? 0 : -ENOMEM;
505 #else
506         return 0;
507 #endif
508 }
509
510 static void __exit exit_nlm(void)
511 {
512         /* FIXME: delete all NLM clients */
513         nlm_shutdown_hosts();
514 #ifdef CONFIG_SYSCTL
515         unregister_sysctl_table(nlm_sysctl_table);
516 #endif
517 }
518
519 module_init(init_nlm);
520 module_exit(exit_nlm);
521
522 /*
523  * Define NLM program and procedures
524  */
525 static struct svc_version       nlmsvc_version1 = {
526                 .vs_vers        = 1,
527                 .vs_nproc       = 17,
528                 .vs_proc        = nlmsvc_procedures,
529                 .vs_xdrsize     = NLMSVC_XDRSIZE,
530 };
531 static struct svc_version       nlmsvc_version3 = {
532                 .vs_vers        = 3,
533                 .vs_nproc       = 24,
534                 .vs_proc        = nlmsvc_procedures,
535                 .vs_xdrsize     = NLMSVC_XDRSIZE,
536 };
537 #ifdef CONFIG_LOCKD_V4
538 static struct svc_version       nlmsvc_version4 = {
539                 .vs_vers        = 4,
540                 .vs_nproc       = 24,
541                 .vs_proc        = nlmsvc_procedures4,
542                 .vs_xdrsize     = NLMSVC_XDRSIZE,
543 };
544 #endif
545 static struct svc_version *     nlmsvc_version[] = {
546         [1] = &nlmsvc_version1,
547         [3] = &nlmsvc_version3,
548 #ifdef CONFIG_LOCKD_V4
549         [4] = &nlmsvc_version4,
550 #endif
551 };
552
553 static struct svc_stat          nlmsvc_stats;
554
555 #define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
556 static struct svc_program       nlmsvc_program = {
557         .pg_prog                = NLM_PROGRAM,          /* program number */
558         .pg_nvers               = NLM_NRVERS,           /* number of entries in nlmsvc_version */
559         .pg_vers                = nlmsvc_version,       /* version table */
560         .pg_name                = "lockd",              /* service name */
561         .pg_class               = "nfsd",               /* share authentication with nfsd */
562         .pg_stats               = &nlmsvc_stats,        /* stats table */
563         .pg_authenticate = &lockd_authenticate  /* export authentication */
564 };